Aluminum template production line discrete event simulation intelligent control method and system

Abstract

The invention provides an intelligent control method and system for discrete event simulation of an aluminum template production line, and relates to the technical field of data processing.The method comprises the steps that a material transfer path instruction and a cutting process key point coordinate sequence are analyzed from a dynamic anti-interference scheduling strategy, the key point coordinate sequence is input into a curve driving algorithm, and a curve driving algorithm is obtained; a parameterized continuous cutting path curve is generated according to the aluminum template contour precision requirement and the cutting head dynamics constraint; the method comprises the following steps: performing equal-bow-height error discretization processing on a continuous cutting path curve to generate a discrete pose point set with a timestamp, superposing AGV dynamic constraints to generate a speed look-ahead control instruction set, splitting the instruction set into independent control units according to process sections, packaging the independent control units into a time trigger message through an industrial bus protocol, and sending the time trigger message to a server; and the simulation clock synchronization signal is injected into a physical execution unit. According to the invention, intelligent and accurate control of the aluminum template production process is realized, and the production efficiency is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an aluminum template production line discrete event simulation intelligent control method and system. BACKGROUND

[0002] As the core component of modern building industrialization, the production quality of aluminum template directly determines the concrete forming precision and construction efficiency. With the upgrading of the building industry, the requirements for light weight, turnover efficiency and recycling rate of the template system are increasing. The traditional aluminum template production line cannot meet the manufacturing requirements of high precision and high efficiency, and an intelligent control system based on real-time data driving needs to be built.

[0003] The traditional technology monitors the welding process by using a single parameter sampling mode, such as collecting the welding current effective value through a current sensor or measuring the temperature through a thermocouple single point. This mode cannot realize the synchronous collection of the welding current pulse waveform (including the time sequence characteristics of the base value and peak current), the infrared temperature field distribution and the vibration spectrum, and does not establish a correlation analysis mechanism for the pulse-temperature-vibration data stream, resulting in an increase in the lag time of welding defect identification and a decrease in the process accuracy. SUMMARY

[0004] The technical problem to be solved by the present application is to provide an aluminum template production line discrete event simulation intelligent control method and system, which realizes intelligent and accurate control of the aluminum template production process and effectively improves the production efficiency.

[0005] To solve the above technical problems, the technical scheme of the present application is as follows: In a first aspect, an aluminum template production line discrete event simulation intelligent control method is provided, which comprises: Step 1: Collecting welding current pulse waveform, infrared temperature field distribution and vibration spectrum data; generating a dynamic event package with a process label according to the pulse-temperature-vibration data stream; and injecting the dynamic event package into the priority queue of the discrete event simulation engine; Step 2: Based on the dynamic event package in the priority queue, analyzing the thermal cumulative gradient parameter in the feature vector; based on the thermal cumulative gradient parameter, calculating the pose compensation amount of the welding robot twin body, and updating the kinematic model parameter of the twin body; Step 3: Based on the updated kinematic model parameter, advancing the discrete event simulation clock; if the simulation event triggers the logistics decision node, suspending the event scheduler and intercepting the thermal constraint state matrix; using the DQN algorithm to generate a dynamic anti-disturbance scheduling strategy based on the thermal constraint state matrix; Step 4, parse the material transfer path instructions and the key point coordinate sequence of the cutting process from the dynamic disturbance-resistant scheduling strategy, input the key point coordinate sequence into the curve driving algorithm, and generate a parameterized continuous cutting path curve according to the aluminum template contour accuracy requirements and the cutting head dynamics constraints; Step 5, discretize the continuous cutting path curve with equal-arc-height error to generate a discrete pose point set with time stamp, superimpose the AGV dynamics constraints to generate a speed look-ahead control instruction set, split the instruction set into independent control units according to the process section, encapsulate it into time-triggered messages through the industrial bus protocol, and inject it into the physical execution unit based on the simulation clock synchronization signal.

[0006] Further, collect welding current pulse waveform, infrared temperature field distribution and vibration spectrum data; generate dynamic event packages with process labels according to the pulse-temperature-vibration data stream; inject the dynamic event packages into the priority queue of the discrete event simulation engine, including: Step 11, generate a three-dimensional sensing signal set by synchronously collecting welding current pulse waveform, infrared temperature field distribution and vibration spectrum data, and aligning the three data with the unified clock source timestamp; Step 12, extract current waveform variation characteristics, temperature heat field gradient characteristics and vibration energy accumulation characteristics based on the three-dimensional sensing signal set, match the process type identifier according to the feature combination, and generate dynamic event entities; Step 13, based on the dynamic event entity, parse the process type identifier to determine the basic priority level, superimpose the feature intensity parameter to calculate the final priority, load the event entity into the priority queue of the discrete event simulation engine in descending order of priority, and clear the historical event entity with the lowest priority when the queue is full.

[0007] Further, the step 2 includes: Step 21, extract the dynamic event package from the priority queue to generate a feature analysis result containing the process label and the temperature field distribution data; based on the temperature field distribution data in the feature analysis result, calculate the heat accumulation gradient change rate between adjacent heat analysis nodes in the welding area to obtain a gradient change rate matrix; Step 22, according to the gradient change rate matrix and the process label, match the pose compensation rules in the process knowledge base, output the compensation rule instruction set; based on the compensation rule instruction set, generate a six-degree-of-freedom pose compensation vector of the welding robot twin; Step 23, parse the pose compensation vector to obtain decoupled translation and rotation components, execute position compensation based on the translation component to generate compensated position coordinates, and execute pose fusion based on the rotation component to generate compensated pose data; based on the compensated pose data, check the joint limits through kinematics inverse solution and generate an out-of-limit state flag; Step 24, scale the compensation amount based on the state flag and the pose compensation vector to generate the final pose parameter, and write the final pose parameter into the kinematics model database to generate the updated kinematics parameter.

[0008] Further, based on the updated kinematics model parameter, advance the discrete event simulation clock; if the simulation event triggers the logistics decision node, suspend the event scheduler and intercept the hot constraint state matrix; use the DQN algorithm to generate the dynamic disturbance-resistant scheduling strategy based on the hot constraint state matrix, including: Step 31, advance the discrete event simulation clock based on the updated kinematics model parameter to generate the current simulation event set; Step 32, receive the simulation event set and parse the event type identifier and device state parameter, generate a decision trigger flag based on the event type identifier matching the logistics decision node trigger condition, and suspend the event scheduler and lock the simulation clock timestamp in response to the flag; Step 33, based on the locked timestamp, intercept the temperature field distribution matrix, thermal cumulative gradient vector and device cooling state flag from the digital twin database, and encapsulate the hot constraint state data into a structured hot constraint state matrix; Step 34, receive the hot constraint state matrix, perform standardization preprocessing to generate an input state vector; input the input state vector into the strategy network to generate an action value evaluation matrix through multi-layer neuron calculation; based on the action value evaluation matrix, perform value-led decision to generate a dynamic disturbance-resistant scheduling strategy.

[0009] Further, parse the material transfer path instruction and cutting process key point coordinate sequence from the dynamic disturbance-resistant scheduling strategy, and input the key point coordinate sequence into the curve driving algorithm to generate a parameterized continuous cutting path curve according to the aluminum template contour accuracy requirement and cutting head dynamics constraint, including: Step 41, parse the dynamic disturbance-resistant scheduling strategy to obtain the material transfer path topology sequence, and separate the cutting process key point coordinate data set; Step 42, based on the key point coordinate data set, construct a path control structure in the process order; in the path control structure, connect the key points by applying the parametric spline algorithm to obtain an initial path curve; scan and extract the contour feature points along the initial path curve, and identify the curvature mutation area; according to the aluminum template contour accuracy constraint, apply smoothing processing in the curvature mutation area to generate an optimized path curve; insert a curvature continuous transition section at the intersection of the path curve to obtain an industrial-grade smooth path curve; Step 43, receive the industrial-grade smooth path curve, parse the cutting head dynamics parameter constraint, calculate the acceleration sensitive area along the path curve, and insert a velocity smoothing section according to the dynamics constraint to generate a position-velocity-acceleration three-order coupled parameterized continuous cutting path curve.

[0010] Further, the continuous cutting path curve is discretized with equal error to generate a discrete pose point set with time stamp, the AGV dynamics constraint is superimposed to generate a speed look-ahead control instruction set, the instruction set is split into independent control units according to the process section, the industrial bus protocol is used to package the time-triggered message, and the physical execution unit is injected based on the simulation clock synchronization signal, including: Step 51, receiving a parameterized continuous cutting path curve, discretizing sampling based on equal error constraint, generating a discrete pose point set; Step 52, receiving the discrete pose point set, analyzing the space coordinates and time stamp information in the pose point, calculating the path curvature distribution based on the space coordinates, identifying the high dynamic constraint section; applying AGV dynamics constraint in the high dynamic constraint section, generating speed profile through speed look-ahead algorithm; fusing the speed profile and the pose point set to generate the motion control instruction set; Step 53, splitting the motion control instruction set into independent control units according to the process characteristics; packaging the control units into time-triggered messages through the industrial bus protocol; injecting the time-triggered messages into the physical execution unit based on the discrete event simulation clock synchronization signal.

[0011] Further, the discrete pose point set is received, the space coordinates and time stamp information in the pose point are analyzed, the path curvature distribution is calculated based on the space coordinates, and the high dynamic constraint section is identified; the AGV dynamics constraint is applied in the high dynamic constraint section, and the speed profile is generated through the speed look-ahead algorithm; the speed profile and the pose point set are fused to generate the motion control instruction set, including: Step 61, based on the space coordinate information of the discrete pose point set, calculating the instantaneous curvature value of the path segment, generating a curvature distribution sequence, identifying and obtaining high dynamic feature points by comparing with a preset curvature threshold, and merging the continuous high dynamic feature points to generate a high dynamic constraint section; Step 62, based on the high dynamic constraint section, combining the AGV maximum centripetal acceleration parameter to calculate the upper limit of the safe speed of each section, iteratively calculating the allowed speed value between adjacent pose points from the end point to the starting point of the section, and applying the AGV maximum acceleration and deceleration constraint to generate a speed decreasing sequence, and generating a speed profile after fusing the uniform speed profile; Step 63, binding the speed value of the speed profile according to the pose point, combining the space coordinates, attitude angle, speed value and time stamp to generate a structured instruction unit, and generating a motion control instruction set after arranging in ascending order according to the time stamp.

[0012] The second aspect is a discrete event simulation intelligent control system for an aluminum template production line, including: The acquisition module is used for acquiring welding current pulse waveform, infrared temperature field distribution and vibration spectrum data; generating dynamic event packets with process labels according to pulse-temperature-vibration data stream; injecting the dynamic event packets into the priority queue of the discrete event simulation engine; The computing module is configured to parse the thermal accumulation gradient parameter in the feature vector based on the dynamic event package in the priority queue, and calculate the pose compensation amount of the welding robot twin based on the thermal accumulation gradient parameter, and update the kinematics model parameter of the twin; The execution module is configured to advance the discrete event simulation clock based on the updated kinematics model parameter, suspend the event scheduler and intercept the thermal constraint state matrix if the simulation event triggers the logistics decision node, and generate a dynamic disturbance-resistant scheduling strategy by using the DQN algorithm on the thermal constraint state matrix. The constraint module is configured to parse the material transfer path instruction and the coordinate sequence of the key points of the cutting process from the dynamic disturbance-resistant scheduling strategy, and input the coordinate sequence of the key points into a curve driving algorithm to generate a parameterized continuous cutting path curve according to the aluminum template contour accuracy requirement and the cutting head dynamics constraint generation parameter. The processing module is configured to perform equal-bow-height-error discretization processing on the continuous cutting path curve to generate a discrete pose point set with a timestamp, superimpose AGV dynamics constraints to generate a speed look-ahead control instruction set, split the instruction set into independent control units according to the process section, encapsulate the instruction set into time-triggered messages through an industrial bus protocol, and inject the physical execution unit based on the simulation clock synchronization signal.

[0013] In a third aspect, a computing device includes: one or more processors; a memory device storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the method.

[0014] In a fourth aspect, a computer-readable storage medium stores a program, which when executed by a processor, implements the method.

[0015] The above-mentioned scheme of the present application at least has the following beneficial effects: By using the multi-source data mechanism of synchronously collecting welding current pulse waveform, infrared temperature field distribution and vibration spectrum, the problem of single parameter sampling and asynchronous data in the traditional technology is solved. By generating a dynamic event package with a process label, the correlation analysis of pulse-temperature-vibration features is realized, the process state perception accuracy is improved, the data sampling interval is shortened, the welding defect identification response speed is accelerated, a high-precision data source is provided for subsequent simulation and decision-making, and the input error of the discrete event simulation engine is effectively reduced.

[0016] The dynamic anti-interference scheduling strategy is generated by intercepting the thermal constraint state matrix and using the DQN algorithm, the problem that the traditional AGV scheduling ignores the thermal state workpiece characteristics is solved, the hot state workpiece transfer rate is reduced, the quality defects caused by thermal deformation are reduced, the response speed to equipment fault disturbance is accelerated, the anti-interference ability is improved, and the effective operation time proportion of the production line is improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flow chart of an aluminum template production line discrete event simulation intelligent control method provided by an embodiment of the application.

[0018] Figure 2 is a schematic diagram of an aluminum template production line discrete event simulation intelligent control system provided by an embodiment of the application. DETAILED DESCRIPTION

[0019] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, the embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0020] As Figure 1 shown, an embodiment of the application proposes an aluminum template production line discrete event simulation intelligent control method, which comprises the following steps: Step 1, collect welding current pulse waveform, infrared temperature field distribution and vibration spectrum data; generate dynamic event package with process label according to pulse-temperature-vibration data stream; inject dynamic event package into the priority queue of the discrete event simulation engine; Step 2, based on the dynamic event package in the priority queue, analyze the thermal accumulation gradient parameters in the feature vector; based on the thermal accumulation gradient parameters, calculate the pose compensation amount of the welding robot twin, and update the kinematic model parameters of the twin; Step 3, based on the updated kinematic model parameters, advance the discrete event simulation clock; if the simulation event triggers the logistics decision node, suspend the event scheduler and intercept the thermal constraint state matrix; use the DQN algorithm to generate a dynamic anti-interference scheduling strategy from the thermal constraint state matrix; Step 4, parse the material transfer path instructions and cutting process key point coordinate sequence from the dynamic anti-interference scheduling strategy, and input the key point coordinate sequence into the curve driving algorithm, generate a parameterized continuous cutting path curve according to the aluminum template contour accuracy requirement and cutting head dynamics constraint; Step 5, the continuous cutting path curve is subjected to equal-arch-height error discretization processing to generate a discrete pose point set with a time stamp, superposition AGV dynamics constraint is generated to generate a speed look-ahead control instruction set, the instruction set is split into independent control units according to process sections, encapsulated into time-triggered messages through an industrial bus protocol, and injected into a physical execution unit based on a simulation clock synchronization signal.

[0021] In the embodiment of the present application, by adopting the multi-source data mechanism of synchronously collecting welding current pulse waveform, infrared temperature field distribution and vibration spectrum, the problem of single parameter sampling and asynchronous data in traditional technology is solved. By generating a dynamic event package with a process label, the correlation analysis of pulse-temperature-vibration characteristics is realized, the process state perception accuracy is improved, the data sampling interval is shortened, the welding defect identification response speed is accelerated, a high-precision data source is provided for subsequent simulation and decision-making, and the input error of the discrete event simulation engine is effectively reduced.

[0022] By intercepting the thermal constraint state matrix and adopting the DQN algorithm to generate a dynamic anti-disturbance scheduling strategy, the problem that traditional AGV scheduling ignores the thermal state workpiece characteristics is solved, the thermal state workpiece transfer rate is reduced, the quality defects caused by thermal deformation are reduced, the response speed to equipment fault disturbance is accelerated, the anti-disturbance ability is improved, and the effective operation time proportion of the production line is improved.

[0023] In a preferred embodiment of the present application, step 1, welding current pulse waveform, infrared temperature field distribution and vibration spectrum data are collected; according to the pulse-temperature-vibration data stream, a dynamic event package with a process label is generated; the dynamic event package is injected into the priority queue of the discrete event simulation engine, including: Step 11, by synchronously collecting welding current pulse waveform, infrared temperature field distribution and vibration spectrum data, three-dimensional sensing signal sets are generated after aligning the time stamps of the three data according to a unified clock source; Step 12, based on the three-dimensional sensing signal set, current waveform change characteristics, temperature thermal field gradient characteristics and vibration energy aggregation characteristics are extracted, the process type identifier is matched according to the feature combination, and a dynamic event entity is generated; Step 13, based on the dynamic event entity, the process type identifier is analyzed to determine the basic priority level, the final priority is calculated by superimposing the feature intensity parameters, the event entity is loaded into the priority queue of the discrete event simulation engine in descending order of priority, and the historical event entity with the lowest priority is cleared when the queue is full.

[0024] The above step 11 uses a high-precision industrial clock source (time accuracy ≤10 μs) to realize the synchronous acquisition of the welding current pulse waveform, the infrared temperature field distribution, and the vibration spectrum data; the welding current pulse waveform is collected by a high-frequency current sensor, the sampling frequency is set to 50 kHz-200 kHz, and the base current, the peak current, and the pulse interval timing characteristics are recorded; the infrared temperature field distribution is collected by an infrared thermal imager, the resolution is 640×512-1280×1024 pixels, the temperature measurement range covers -20℃-500℃, and the spatial distribution matrix of the output temperature field is recorded; the vibration spectrum data is collected by a three-axis acceleration sensor, the sampling frequency is 2 kHz-10 kHz, and the vibration frequency range covers 1 Hz-5 kHz.

[0025] After the three-way data acquisition, the time stamps of the unified clock source (time stamp interval ≤1 ms) are aligned, and a three-dimensional sensing signal set containing the time dimension, the signal type dimension, and the characteristic value dimension is generated, and the time synchronization error is controlled within ±50 μs.

[0026] The above step 12 extracts key features from the three-dimensional sensing signal set: the current waveform variation characteristics include the peak current fluctuation amplitude (the value range is ±5%-±20% of the rated peak value) and the pulse width variation rate (the value range is ±10%-±50%); the temperature thermal field gradient characteristics include the maximum temperature gradient direction (0°-360° spatial angle) and the gradient variation rate (0.5 K / mm·s-5 K / mm·s); the vibration energy aggregation characteristics include the main vibration frequency band (100 Hz-5 kHz) and the proportion of the energy of the frequency band in the total energy (30%-90%).

[0027] The extracted feature combination is compared with the preset process feature library, and the corresponding process type identifier (such as “normal welding”, “spatter warning”, and “thermal deformation risk”) is matched, each identifier is associated with a specific process parameter range, and a dynamic event entity containing the characteristic value, the time stamp, and the process identifier is generated accordingly.

[0028] The above step 13 determines the basic priority level based on the process type identifier of the dynamic event entity: the core process abnormality class (such as “unstable molten pool”) has a basic priority of 5 levels (the highest), the process fluctuation class (such as “small amplitude current fluctuation”) has a basic priority of 3-4 levels, and the normal state class has a basic priority of 1-2 levels.

[0029] The feature intensity parameter is used to calculate the final priority, and the feature intensity parameter is valued according to the degree of deviation of the characteristic value from the normal range (0.1-1.0), wherein the serious deviation is 0.8-1.0, and the slight deviation is 0.1-0.3.

[0030] The final priority is the weighted result of the basic priority and the feature intensity parameter (for example, the priority of the core abnormality class can be increased to 5.0-5.5 after superimposing the high intensity parameter).

[0031] The priority queue capacity of the discrete event simulation engine is set to 50-200 event entities, loaded in descending order of final priority; when the queue is full, 1-5 historical event entities with the lowest priority are automatically cleared (events with the earliest timestamp and priority ≤ 2 are preferentially cleared).

[0032] In the embodiment of the application, the synchronous acquisition mechanism ensures that the time deviation of the three-way data is extremely small, avoids feature misjudgment caused by asynchronous data, and improves the accuracy of process state perception; the feature extraction and process identification matching realize accurate conversion of data to events, so that virtual simulation can reflect the subtle changes of physical production in real time; the priority dynamic sorting mechanism ensures that key process events (such as abnormal early warning) are preferentially processed by the simulation engine, avoids important information lag, and at the same time prevents data overflow through queue capacity management, thereby improving simulation efficiency.

[0033] In a preferred embodiment of the application, step 2, based on the dynamic event package in the priority queue, the thermal accumulation gradient parameter in the feature vector is analyzed; based on the thermal accumulation gradient parameter, the pose compensation amount of the welding robot twin is calculated, and the kinematic model parameters of the twin are updated, including: Step 21, extracting the dynamic event package from the priority queue, generating a feature analysis result containing the process label and the temperature field distribution data; based on the temperature field distribution data in the feature analysis result, the thermal accumulation gradient change rate between adjacent thermal analysis nodes in the welding area is calculated, and a gradient change rate matrix is obtained; Step 22, according to the gradient change rate matrix and the process label, matching the pose compensation rule in the process knowledge base, outputting a compensation rule instruction set; based on the compensation rule instruction set, generating a six-degree-of-freedom pose compensation vector of the welding robot twin; Step 23, analyzing the pose compensation vector to obtain decoupled translation and rotation components, executing position compensation based on the translation component to generate compensated position coordinates, and executing pose fusion based on the rotation component to generate compensated pose data; based on the compensated pose data, checking the joint limit through inverse kinematics and generating an out-of-limit state flag; Step 24, based on the state flag and the pose compensation vector, scaling the compensation amount to generate final pose parameters, and writing the final pose parameters into the kinematic model database to generate updated kinematic parameters.

[0034] Step 21: The highest priority dynamic event package is extracted from the priority queue of the discrete event simulation engine, and the feature analysis result containing the "welding stage", "material thickness" process label and infrared temperature field distribution data (spatial resolution of 640x512 pixels, temperature measurement range of 25-300°C) is parsed. Based on the temperature field data, a 10x10 thermal analysis node grid is divided in the welding area (usually 50mm on both sides of the weld center), and the node spacing is set to 5mm.

[0035] The ratio of the temperature difference between adjacent nodes to the spatial distance is calculated to obtain the thermal accumulation gradient change rate, forming a 10x10 gradient change rate matrix. In aluminum template welding, the reasonable range of gradient change rate is 1K / mm-20K / mm, and when it exceeds 20K / mm, it is marked as "high thermal gradient area".

[0036] Step 22: The gradient change rate matrix and the process label are input into the process knowledge base for matching. The process knowledge base contains pose compensation rules for different aluminum template thicknesses (2-10mm) and welding currents (100-300A), for example, when the process label is "5mm thick aluminum template + 200A current" and the gradient change rate is >15K / mm, the "enhanced compensation rule" is matched. Based on the matched compensation rule instruction set, a six-degree-of-freedom pose compensation vector (x, y, z translation components and roll, pitch, yaw rotation components) is generated, where the translation component value range is -0.5mm-+0.5mm, and the rotation component value range is -0.5°-0.5°.

[0037] Step 23: The pose compensation vector is decoupled to separate the x / y / z axis translation components and the rotation components around each axis; the translation components are directly superimposed on the current position coordinates of the welding robot twin body to generate the compensated position data; the rotation components are fused with the current pose data through a quaternion fusion algorithm to generate the compensated pose data; based on the robot joint structure parameters (each rotation joint angle range -180°-180°), the joint angles corresponding to the compensated pose are calculated through inverse kinematics, and if a joint angle exceeds -170° or 170° (10° safety margin is reserved), a "over-limit state flag" is generated.

[0038] Step 24: If no over-limit state flag is generated, the compensated pose data is directly used as the final pose parameter; if there is an over-limit flag, the pose compensation vector is scaled by 0.8-0.9 according to the over-limit degree (the proportion of over-limit to safety margin), and the over-limit is eliminated by recalculating and verifying; the final pose parameter is written into the kinematics model database, covering the original D-H parameters and joint zero point offset model parameters, and the real-time update of the twin body kinematics model is completed.

[0039] Construction, training and implementation process of welding robot twin kinematics model: Model construction: Take the actual welding robot (6-axis serial structure) as the physical prototype, obtain the robot body structure size through three-dimensional laser scanning, and establish a 1:1 digital model combined with CAD drawings; based on the Denavit-Hartenberg method, construct the kinematics model, define the joint coordinate system (joints 1-6 correspond to base rotation, large arm swing, small arm swing, wrist rotation, wrist swing, and end rotation), initial parameters include joint link length (500mm-1200mm), joint offset (0mm-50mm), and joint angle range (-180°-180°), and are stored in the model database.

[0040] Model training: Under the conditions of robot empty load and load (5kg-20kg), control the robot to move according to the preset trajectory (including straight line and circular arc typical path), collect actual pose data through the laser tracker (accuracy ±0.01mm) installed at the end, record joint encoder data synchronously, and accumulate 1000+ groups of motion data; compare the deviation between twin simulation pose and actual pose, optimize D-H parameters by least squares method, and focus on correcting joint zero offset and link length error, so that the static pose error is reduced from the initial ±0.3mm to within ±0.05mm.

[0041] Detailed calculation process and value range of least squares method to optimize D-H parameters: Before optimization, the initial D-H parameters of the welding robot twin need to be determined, including joint angle (θ), link length (a), link offset (d) and joint offset (α); combined with the structural characteristics of the aluminum template welding robot, the initial value range is as follows: the initial value of joint angle θ is set according to the robot zero pose, ranging from -180° to 180°; the link length a is determined according to the actual size of the mechanical arm, ranging from 50mm to 500mm; the link offset d corresponds to the joint axial distance, ranging from 0mm to 200mm; the joint offset α is the angle between adjacent link coordinate systems, ranging from -90° to 90°; the initial parameters are obtained through the robot factory calibration manual, with an error controlled within ±0.5mm (length parameter) and ±0.5° (angle parameter).

[0042] 30-50 groups of pose data under typical welding conditions are extracted from the dynamic event package as a sample set, each group of data containing compensated pose data (actual measurement value) and theoretical pose value calculated by the current D-H parameters; the sample points need to cover the main area of the robot workspace, including the near origin area, the edge area of the working radius and the common welding pose area, to ensure the uniformity of data distribution; the sample data is preprocessed to eliminate abnormal values (deviation more than 3 times the standard deviation of the mean value) caused by sensor noise, and the effective sample size is not less than 90% of the initial sample set.

[0043] The deviation between the actual measured pose and the theoretical pose is taken as the optimization target, and the deviation includes position deviation (Δx, Δy, Δz) and attitude deviation (Δα, Δβ, Δγ); the unit of position deviation is mm, and the unit of attitude deviation is degree; the least square method adjusts the D-H parameters through iteration to minimize the overall deviation square sum; in each iteration, the parameters that have significant impact on the deviation are adjusted first: the link length a and the offset d mainly affect the position deviation, and when the position deviation exceeds 0.3 mm, they are mainly optimized; the joint angle θ and the offset α mainly affect the attitude deviation, and when the attitude deviation exceeds 0.3°, they are mainly adjusted.

[0044] To avoid parameter optimization beyond the mechanical structure limit, set the D-H parameter adjustment range: the single adjustment amount of joint angle θ is not more than 5°, and the cumulative adjustment amount is not more than 10°; the single adjustment amount of link length a and offset d is not more than 0.5 mm, and the cumulative adjustment amount is not more than 2 mm; the single adjustment amount of joint offset α is not more than 3°, and the cumulative adjustment amount is not more than 5°; in the iteration process, if the deviation square sum changes less than 0.01 for 3 consecutive iterations, stop iteration and output the optimized D-H parameters; the optimization process needs to be completed within 100 ms to ensure real-time performance.

[0045] Substitute the optimized D-H parameters into the kinematics model to recalculate the theoretical pose of the sample set, verify whether the position deviation is controlled within 0.1 mm and the attitude deviation is controlled within 0.1°; if the verification fails, the number of sample points needs to be increased (to 60-80 groups) or the iteration stopping condition needs to be relaxed (change less than 0.02), and the optimization process needs to be executed again until the accuracy requirement is met.

[0046] Dynamic compensation training: introduce a thermal deformation influence factor, collect robot pose drift data under different welding currents and temperature fields, establish a mapping relationship between thermal cumulative gradient and pose deviation, embed the relationship into the model compensation module, and complete the dynamic error correction training.

[0047] Model implementation: based on Unity3D to build a twin visualization platform, and communicate with the physical robot control system through OPCUA protocol; receive the joint state data of the physical robot in real time, drive the twin to move synchronously; at the same time, inject the updated kinematics parameters generated in step 24 into the twin model to realize the dynamic adaptation of the virtual model and the physical entity; through the edge computing module (response time < 50ms), complete the pose compensation calculation and model parameter update, and ensure the real-time mapping of the twin motion state and the actual welding process.

[0048] In the embodiment of the present application, through the pose compensation driven by the heat accumulation gradient, the pose deviation of the twin and the physical robot is significantly reduced, and the consistency of virtual-real mapping is improved. The dynamic correction is carried out for the robot pose drift caused by welding heat accumulation, which effectively reduces the influence of thermal deformation on the welding quality. Through joint limit checking and compensation amount scaling, the risk of joint over-limit in the robot motion process is avoided, and the equipment operation safety is improved. Based on real-time data, the kinematics parameters are dynamically updated, so that the twin model can adapt to the working condition changes under different welding processes and material thickness, and the application range of the model is expanded.

[0049] In a preferred embodiment of the present application, step 3, based on the updated kinematics model parameters, the discrete event simulation clock is advanced; if the simulation event triggers the logistics decision node, the event scheduler is suspended and the heat constraint state matrix is intercepted; the heat constraint state matrix is generated by using the DQN algorithm to generate a dynamic anti-disturbance scheduling strategy, including: Step 31, based on the updated kinematics model parameters, the discrete event simulation clock is advanced, and the current simulation event set is generated; Step 32, receive the simulation event set and analyze the event type identifier and the device state parameter, generate a decision trigger flag based on the event type identifier matching the logistics decision node trigger condition, and respond to the flag to suspend the event scheduler and lock the simulation clock timestamp; Step 33, based on the locked timestamp, the temperature field distribution matrix, the heat accumulation gradient vector and the device cooling state marker are intercepted from the digital twin database, and the heat constraint state data is packaged into a structured heat constraint state matrix; Step 34, receive the heat constraint state matrix, perform standardization preprocessing to generate an input state vector; input the input state vector into the strategy network, calculate the action value evaluation matrix through multiple layers of neurons, and execute the value leading decision based on the action value evaluation matrix to generate a dynamic anti-disturbance scheduling strategy.

[0050] Step 31: Based on the updated kinematic model parameters (including optimized D-H parameters of joint angles and link lengths), the discrete event simulation clock is advanced at a fixed time step. The time step is set according to the process precision requirement, with a value range of 10-50 ms, ensuring that the simulation rhythm is consistent with the time scale of the physical production process. With each time step advanced, the virtual state of each device in the simulation environment is updated synchronously, generating a current simulation event set containing device operating status, process progress, and abnormal warning information.

[0051] Step 32: After receiving the simulation event set, the event type identifier and device state parameters (such as AGV current position, welding robot working status, workpiece temperature) are parsed. Based on the pre-set logistics decision node trigger conditions (such as workpiece welding completion and temperature dropping to a pre-set threshold, certain work station material inventory below a safety value), matching is performed to generate a decision trigger flag. When the flag is "trigger", the event dispatcher is suspended immediately, and the current simulation clock timestamp is locked, with a precision controlled within ±1 ms.

[0052] Step 33: Based on the locked timestamp, the thermal constraint state data at the corresponding time from the digital twin database is extracted: the temperature field distribution matrix is stored in a two-dimensional grid form, with a grid size of 50x50-200x200 pixels, and a temperature value range of 25-300℃, with an accuracy of integer bits; the thermal cumulative gradient vector is generated according to the sub-regions divided by the welding area, with each sub-region corresponding to a gradient value, and a value range of 5-30 K / mm; the device cooling state flag is encoded in binary (0 represents no cooling started, and 1 represents cooling in progress); the above data is encapsulated as a structured thermal constraint state matrix according to the hierarchical structure of "temperature field-gradient-cooling state", with the matrix dimension dynamically adjusted according to the production scale, usually 100x100-500x500.

[0053] Step 34: Standardized preprocessing is performed on the thermal constraint state matrix, mapping the temperature value and gradient value to the 0-1 interval respectively to eliminate dimensional differences and generate an input state vector; the input vector is input into the pre-trained policy network, and the network generates an action value evaluation matrix through multi-layer neuron calculation (the number of hidden layers is 3-5, and the number of neurons in each layer is 64-256), with the matrix elements representing the expected value of different scheduling actions (such as AGV path selection, transfer timing adjustment, and device load distribution), and the value range being -10 to 10; based on the highest value element in the matrix, a value-dominated decision is made to generate a dynamic anti-disturbance scheduling strategy containing material transfer priority, path planning direction, and device coordination sequence.

[0054] DQN model construction, training, and implementation process: Model construction: The deep Q network (DQN) architecture is adopted, the input layer dimension is consistent with the flattened dimension of the thermal constraint state matrix (e.g. a 100x100 matrix corresponds to a 10000-dimensional input); the hidden layer uses a fully connected structure to avoid the problem of gradient disappearance; the output layer dimension matches the action space dimension (usually 10-50 dimensions), and the output value represents the Q value (action value) of each action.

[0055] An experience replay pool with a capacity of 10000-100000 samples is constructed, each sample contains a "state-action-reward-next state" four-tuple, which is used to store historical interaction data, and the samples are shuffled and reused; the target network and policy network are introduced in parallel, the target network parameters are updated every 1000-5000 steps, and the parameter copying method of the policy network is used to improve the stability of training.

[0056] Model training: Collect the production log data of the past 6 months, including the thermal constraint state, scheduling action and corresponding production indicators (such as transfer time, thermal deformation, equipment utilization) under different working conditions, construct an initial training data set, and the sample size is not less than 50000; the reward value is calculated by weighting multiple dimensions, including: thermal workpiece transfer rate reduction (weight 0.3), AGV path length shortening (weight 0.2), equipment idle time reduction (weight 0.2), production cycle compliance rate (weight 0.3), single step reward value range is -5 to 10.

[0057] Training process: Select action (ε initial value is 1.0, linearly decayed to 0.1 with training steps), randomly select 32-128 samples from the experience replay pool each time, and minimize the loss function (mean square error loss) by gradient descent method; the training iteration number is 50000-200000 steps, the learning rate is set to 0.0001-0.001, the model performance is verified every 1000 steps, and the optimal model parameters are saved.

[0058] Specific calculation process of gradient descent method mean square error loss function: Extract batch samples from the training sample set, each sample contains the input features corresponding to the thermal constraint state matrix, the action value output predicted by the model, and the target value obtained by actual interaction; for a single sample, calculate the difference between the predicted action value and the target value, which reflects the deviation of a single prediction, usually the value range is -10 to 10 (depending on the reward value setting); square the difference, the purpose is to amplify the influence of large errors and eliminate the problem of positive and negative deviation offset, the squared value range is generally 0 to 100; add the squared errors of all samples in the batch, and then divide by the number of samples to get the batch average loss value, the mean square error loss value is usually controlled between 0 and 50, if it exceeds 50, it means that the model prediction deviation is large.

[0059] With the mean square error loss value as the optimization goal, the partial derivative of the loss function to each weight parameter in the strategy network, that is, the gradient, is calculated, the gradient direction represents the direction in which the loss function increases, the value range changes with the parameter dimension and the network depth, and is usually between-0.1 and 0.1; the parameter adjustment direction is determined according to the gradient direction (opposite to the gradient direction), and the adjustment amplitude is controlled in combination with a preset learning rate; the learning rate is generally set to be between 0.001 and 0.01; if the learning rate is too small, the convergence is slow, and more than 1000 iterations are required to stabilize; if the learning rate is too large, parameter oscillation may be caused, and the optimal solution cannot be converged.

[0060] In each iteration, the parameter update amount is obtained by multiplying the gradient by the learning rate, and the network weight is fine-tuned; after completing a parameter update, a new batch of samples is extracted to calculate the loss value, and it is judged whether the loss decreases; if the loss does not decrease significantly (the decrease amplitude is less than 0.001) for 30 iterations, the learning rate is dynamically reduced (usually reduced to 1 / 10 of the original); the entire training process requires 5000 to 10000 iterations until the loss value is stabilized between 0 and 5, at which time the model prediction accuracy meets the scheduling strategy generation requirement.

[0061] Model implementation: The model is deployed on an industrial server (CPU is Intel Xeon E5-2690v4, and GPU is NVIDIA Tesla P100), supports parallel computing acceleration, and the inference time of a single state vector is controlled to be within 50ms; a communication interface between the developed model and the discrete event simulation engine is developed, a TCP / IP protocol is used to realize state data input and scheduling strategy output, and the data transmission delay is ≤10ms; an online model update switch is set, and incremental training (1000-5000 steps) is performed every 24 hours using the production data of the day to ensure that the model adapts to changes in the production environment.

[0062] In the embodiment of the application, by intercepting the thermal constraint state matrix and integrating the DQN algorithm, the scheduling strategy fully considers the thermal state characteristics of the workpiece, avoids quality defects caused by thermal deformation, and reduces invalid material transfer. The suspended event scheduler realizes real-time linkage of simulation and decision-making, and the strategy network trained by reinforcement learning improves the response speed of equipment failure and maintains the stability of the production line.

[0063] In a preferred embodiment of the application, step 4, the material transfer path instruction and the cutting process key point coordinate sequence are parsed from the dynamic anti-interference scheduling strategy, and the key point coordinate sequence is input into the curve driving algorithm to generate a parameterized continuous cutting path curve according to the aluminum template contour accuracy requirement and the cutting head dynamics constraint, including: Step 41, the material transfer path topology sequence is parsed from the dynamic anti-interference scheduling strategy, and the cutting process key point coordinate data set is separated; Step 42: Based on the key point coordinate dataset, construct the path control structure according to the process sequence; within the path control structure, apply the parametric spline algorithm to connect the key points to obtain the initial path curve; scan along the initial path curve to extract contour feature points and identify curvature change regions; according to the aluminum template contour accuracy constraints, apply smoothing processing to the curvature change regions to generate optimized path curves; insert curvature continuous transition segments at the intersection of path curves to obtain industrial-grade smooth path curves. Step 43: Receive the industrial-grade smooth path curve, analyze the dynamic parameter constraints of the cutting head, calculate the acceleration sensitive region along the path curve, insert a velocity smoothing segment according to the dynamic constraints, and generate a parameterized continuous cutting path curve with third-order coupling of position, velocity and acceleration.

[0064] In step 41 above, the structured data of the dynamic anti-disturbance scheduling strategy is parsed, and the coordinates of the starting point, via points, and ending point of the material transfer path are extracted to form a topological sequence. The key point coordinate dataset of the cutting process is separated from the strategy data. The key point coordinates are sorted according to the cutting order. The number of key points of a single aluminum template is usually 8-20, and the coordinate value range matches the size of the aluminum template (generally within a plane range of 0-6000mm × 0-2000mm). Step 42 above, path control structure construction: Based on the sorted key point coordinate dataset, construct a path control structure containing the starting point, turning point and ending point according to the cutting process order (such as first the border and then the reinforcing rib). Each key point is associated with a process attribute label (such as "corner", "straight line segment" and "arc segment").

[0065] The parametric spline algorithm is used to connect adjacent key points to generate an initial path curve; the curve sampling interval is set to 0.5-2mm to ensure that subtle contour changes can be captured.

[0066] Parametric spline algorithm calculation process: Extract three consecutive adjacent key points from the key point coordinate dataset of the cutting process as a group of control points. The spatial spacing of each group of control points should be maintained within the range of 5-50mm (dynamically adjusted according to the aluminum template contour accuracy requirements: 5-15mm for high-precision areas and 20-50mm for low-precision areas); perform coordinate verification on each group of control points and remove abnormal points that exceed the process allowable deviation (±0.3mm). If more than two abnormal points appear consecutively, a re-acquisition mechanism is triggered; assign an independent parameter range to each group of control points, and set the parameter value range to 0-1 (0 for the starting point and 1 for the ending point).

[0067] Convert physical coordinate distances to parameter step sizes, with the step size interval between adjacent parameter points controlled between 0.01 and 0.1: This applies to regions with gentle curvature changes (curvature value < 0.05 mm). -1 Using a large step size of 0.05-0.1, in areas with drastic curvature changes (curvature value ≥ 0.05mm)-1 ) Take small step size of 0.01-0.03 to ensure the curve details in high curvature area.

[0068] Based on the processed control points and parameter intervals, the basic spline curve is constructed; in the curve generation process, the first derivative (slope) of adjacent segments at the connection point needs to be ensured to be continuous, and the derivative difference is controlled within ±0.1; for the right angle transition area (included angle of 85°-95°) of the aluminum template, a transition section of 3-10 mm is extended on both sides of the corner, and the slope change rate of the curve in the transition section is controlled within 0.02-0.1 mm / mm to avoid sharp edge cutting defects. -1

[0069] The initial spline curve is modified by introducing a smoothing factor, and the value range of the smoothing factor is 0.1-0.5 (the larger the value, the smoother the curve): for the template plane area (surface roughness requirement Ra≤1.6 μm), the smoothing factor is 0.3-0.5; for the feature area with patterns or grooves, the smoothing factor is 0.1-0.2 to retain the detail features; after adjustment, the maximum deviation of the curve from the original key point needs to be verified to ensure that it is not more than ±0.05 mm (high precision area) or ±0.1 mm (ordinary area).

[0070] All segmented curves are traversed to check the second derivative (curvature) difference of adjacent curve segments, and the difference is required to be ≤0.02 mm -2 ; if it exceeds the range, a transition spline with a length of 5-20 mm is inserted between the two curves, and the curvature change of the transition spline needs to show a linear decay trend, gradually transitioning from the curvature value of the high curvature segment to the curvature value of the low curvature segment to ensure the smoothness of the global curve.

[0071] The curvature value is calculated point by point along the initial path curve, and when the curvature change of adjacent two points exceeds 0.05-0.2 m -1 (adjust according to the aluminum template contour precision level, take a small value for high precision template), mark the area as a curvature mutation area. The curvature mutation area is fitted by multiple line segments, and the length of each fitting line segment is controlled within 5-50 mm (short line segments are used for high curvature areas, and long line segments are used for low curvature areas); the position accuracy of the original key point is preserved during fitting, and the deviation is controlled within 0.03-0.1 mm. A smooth transition section is inserted between the line segments of the multiple line fitting, and the transition section length is 5-30 mm, so that the tangent angle of adjacent line segments changes from mutation to continuous change (the angle change rate is ≤5° / mm); after optimization, the maximum curvature fluctuation amplitude of the path curve is reduced to 30% or less of the original mutation value, and an industrial-level smooth path curve is generated. ​The step 43 extracts the cutting head dynamics constraint parameters from the cutting device parameter library, including the maximum running speed (usually 500-2000 mm / min), the maximum acceleration (usually 300-1000 mm / s 2 ), and the maximum jerk (usually 5000-20000 mm / s 3 ), to form the constraint boundary conditions; the curvature derivative of each point along the industrial-grade smooth path curve is calculated, and when the absolute value of the curvature derivative exceeds 0.01-0.05 m -2 , the area is marked as an acceleration-sensitive area (such an area is prone to vibration caused by speed changes). In the acceleration-sensitive area, a speed smoothing section is inserted according to the dynamics constraint; the starting speed of the smoothing section is set to 80%-90% of the current path speed, the terminal speed matches the target speed of the next area, the speed change is realized through linear transition, and the transition length is 10-100 mm (a long transition section is taken in a high-curvature area); the position coordinates, speed planning, and acceleration limit data are fused to generate a parameterized continuous cutting path curve with three-order coupling of position-speed-acceleration; the speed of any point in the curve does not exceed 90% of the maximum speed, and the acceleration does not exceed 80% of the maximum acceleration, so as to ensure smooth motion of the cutting head.

[0072] In the embodiment of the application, through curvature smoothing processing and dynamics constraint optimization, the deviation of the cutting path from the aluminum template profile is controlled within ±0.1-±0.5 mm, meeting the high-precision machining requirement. The three-order coupled path curve eliminates the speed and acceleration mutations, reduces the cutting head vibration and equipment impact, and prolongs the service life of mechanical parts. The dynamic anti-interference scheduling strategy is combined with the smooth path to maximize the cutting speed and shorten the machining cycle on the premise of ensuring the accuracy.

[0073] In a preferred embodiment of the application, step 5 performs equal-arc-height-error discretization processing on the continuous cutting path curve to generate a discrete pose point set with time stamps, superimposes AGV dynamics constraints to generate a speed look-ahead control instruction set, splits the instruction set into independent control units according to process sections, encapsulates the instruction set into time-triggered messages based on an industrial bus protocol, and injects the physical execution unit based on a simulation clock synchronization signal, including: Step 51 receives the parameterized continuous cutting path curve, performs discretization sampling based on the equal-arc-height-error constraint, and generates a discrete pose point set; Step 52 receives the discrete pose point set, analyzes the space coordinates and time stamp information in the pose points, calculates the path curvature distribution based on the space coordinates, identifies the high-dynamic constraint section, applies the AGV dynamics constraint in the high-dynamic constraint section, generates a speed profile through a speed look-ahead algorithm, and fuses the speed profile and the pose point set to generate a motion control instruction set; Step 53, according to the process characteristics, the motion control instruction set is divided into independent control units; through the industrial bus protocol, the control units are packaged as time-triggered messages; based on the discrete event simulation clock synchronization signal, the time-triggered messages are injected into the physical execution unit.

[0074] The above step 51, according to the aluminum template process precision grade threshold range, the first precision area (such as the splicing surface) is set to 0.01-0.03mm, the second precision area (such as the non-working surface) is set to 0.03-0.05mm, and the discretization error is ensured to be not more than 60% of the process allowable value; initial sampling is performed along the continuous cutting path curve, and the basic interval is set to 5mm; real-time monitoring of the curve curvature change is performed, and when the curvature is >0.02mm -1 , the sampling interval is automatically reduced to 1-3mm; when the curvature is ≤0.02mm -1 , the interval is expanded to 5-10mm, and the total number of sampling points of a single path is controlled to be 1000-3000.

[0075] Pose information generation: each discrete point contains three-dimensional coordinates (X / Y axis range 0-12m, Z axis range 0-0.6m, coordinate value retained to 0.001mm), attitude angle (rotation angle -180°-180°, accuracy ±0.05°) and time stamp (generated based on simulation clock, time interval 0.005-0.05s, small value in high-curvature area).

[0076] The generated pose point set is calculated for segment arch height error, and if the error of a segment exceeds the threshold, 2-4 supplementary sampling points are inserted in the segment until the error falls within the threshold; after verification, the discrete pose point set with time stamp is output, and the data format uses JSON structure to store coordinate and attitude information.

[0077] The above step 52, based on the X / Y coordinates of the discrete pose points, the curvature values (unit mm -1 ) of adjacent points are calculated by the three-point method, and the calculation interval is consistent with the sampling interval; when the curvature values of the continuous 3 points are all >0.03mm -1 and the distance between adjacent points is <4mm, it is marked as a high-dynamic constraint section (such as a right-angle turn or an arc-shaped groove).

[0078] The AGV device parameter library is called, and the parameter range is constrained: the maximum travel speed is 1.5-2.5m / s in the empty state and 0.8-1.5m / s in the full load state; the maximum acceleration is 0.3-1.0m / s 2 , and the maximum jerk is 8-15m / s 3 ; the turning speed reduction rule is: when the curvature is >0.05mm -1 , the speed is reduced by 50%-70%, and when the curvature is 0.03-0.05mm-1 The speed is reduced by 30%-50%.

[0079] The deceleration program is started at 80-150 mm before the high dynamic section, and the deceleration acceleration is controlled at 0.5-0.8 m / s 2 ; the speed is forcibly limited to 0.5-0.8 m / s when entering the section, and the speed fluctuation amplitude in the section is ≤±0.05 m / s; after leaving the section, it is linearly recovered to the normal driving speed at an acceleration of 0.4-0.6 m / s 2 ; the speed parameters (instantaneous speed, acceleration) are bound with the pose point coordinates and time stamps to generate a set of motion control instructions, each instruction containing control parameters of 15-30 consecutive pose points.

[0080] Model construction process: receive curvature distribution, spatial coordinates, and time stamp data of discrete pose point set; establish control rule library based on AGV dynamics constraints, including deceleration start distance calculation formula, speed limit threshold table, and acceleration adjustment coefficient.

[0081] Output layer: generate speed profile curve, acceleration change rate, and section transition control signal.

[0082] Model training optimization: collect path and speed data of 1000+ actual cutting operations, covering different curvatures and load conditions; adjust the deceleration start distance coefficient (0.8-1.2) and acceleration attenuation factor (0.5-0.8) through trial and error method to minimize speed fluctuations; take actual path tracking error <0.1 mm and speed overshoot <5% as the qualified standard.

[0083] Engineering implementation: integrated into the PLC module of the AGV controller, using a dual-core processor (≥1 GHz) to ensure real-time performance; interact with the cutting robot through EtherCAT bus with a communication period ≤1 ms; built-in parameter anomaly detection module triggers speed reduction protection when the dynamics parameters exceed the safety range.

[0084] The above step 53 is divided into coarse cutting section, fine cutting section, and transition section independent units according to process features, each unit containing 80-300 consecutive instructions, with 200-500 ms buffer time set between units to prevent instruction execution conflicts; ProfinetRT protocol is used, the message frame structure includes instruction ID (1-255), execution timestamp (precision ±5 μs), control parameter area (including speed and position instructions), and CRC32 check code; the high dynamic section message period is 1-3 ms, and the gentle section is 3-10 ms.

[0085] The PTP synchronization signal (synchronization error < 30 mu s) based on the discrete event simulation clock sends the message 10-20 ms before the preset execution time; the physical execution unit (cutting robot, AGV driver) receives the delay and needs to be ≤15 ms, and if the timeout is not received, the retransmission is triggered; the execution unit feeds back the instruction execution state (success / exception code) in real time, the feedback period is 50 ms, and the maximum retransmission is 3 times with an interval of 100 ms.

[0086] In the embodiment of the application, the equal-arch-height discretization controls the pose tracking error within ±0.03 mm, which is improved compared with the traditional uniform sampling, and the qualified rate of the aluminum template splicing gap is improved. The AGV dynamics constraint model reduces the sudden acceleration operation, the drive motor temperature rise is reduced, and the equipment failure rate is reduced.

[0087] As shown in Figure 2 The embodiment of the application also provides an aluminum template production line discrete event simulation intelligent control system, which comprises: A collection module is configured to collect welding current pulse waveform, infrared temperature field distribution and vibration spectrum data; generate a dynamic event package with a process label according to the pulse-temperature-vibration data stream; and inject the dynamic event package into a priority queue of a discrete event simulation engine; A calculation module is configured to analyze thermal accumulation gradient parameters in a feature vector based on the dynamic event package in the priority queue; calculate a pose compensation amount of a welding robot twin based on the thermal accumulation gradient parameters; and update kinematics model parameters of the twin; An execution module is configured to advance a discrete event simulation clock based on the updated kinematics model parameters; if a simulation event triggers a logistics decision node, suspend an event scheduler and intercept a thermal constraint state matrix; and generate a dynamic anti-disturbance scheduling strategy by using the DQN algorithm on the thermal constraint state matrix; A constraint module is configured to analyze material transfer path instructions and cutting process key point coordinate sequences from the dynamic anti-disturbance scheduling strategy; input the key point coordinate sequences into a curve driving algorithm; and generate a parameterized continuous cutting path curve according to the aluminum template contour accuracy requirement and cutting head dynamics constraint generation parameters; A processing module is configured to perform equal-arch-height error discretization processing on the continuous cutting path curve to generate a discrete pose point set with a time stamp; superimpose AGV dynamics constraint to generate a speed look-ahead control instruction set; split the instruction set into independent control units according to process sections; package the instruction set into time-triggered messages through an industrial bus protocol; and inject the time-triggered messages into a physical execution unit based on a simulation clock synchronization signal.

[0088] It should be noted that the system corresponds to the above method, and all implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0089] The embodiment of the present application also provides a computing device, comprising a processor, a memory storing a computer program, the computer program being executed by the processor to perform the method described above. All implementation manners in the method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0090] The embodiment of the present application also provides a computer readable storage medium storing instructions, which, when executed on a computer, cause the computer to perform the method described above. All implementation manners in the method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0091] The above is the preferred embodiment of the present application, and it should be pointed out that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. An intelligent control method for discrete event simulation of an aluminum die plate production line, characterized in that, The method comprises: Step 1, collect welding current pulse waveform, infrared temperature field distribution and vibration spectrum data; according to the pulse-temperature-vibration data stream, generate a dynamic event package with a process label; inject the dynamic event package into the priority queue of the discrete event simulation engine; Step 2, based on the dynamic event package in the priority queue, analyze the thermal accumulation gradient parameters in the feature vector; based on the thermal accumulation gradient parameters, calculate the pose compensation amount of the welding robot twin, and update the kinematic model parameters of the twin; Step 3, based on the updated kinematic model parameters, advance the discrete event simulation clock; if the simulation event triggers the logistics decision node, suspend the event scheduler and intercept the thermal constraint state matrix; use the DQN algorithm to generate a dynamic anti-interference scheduling strategy from the thermal constraint state matrix; Step 4, parse the material transfer path instructions and cutting process key point coordinate sequence from the dynamic anti-interference scheduling strategy, and input the key point coordinate sequence into the curve driving algorithm; according to the aluminum template contour accuracy requirement and the cutting head dynamics constraint, generate a parameterized continuous cutting path curve; Step 5, discretize the continuous cutting path curve with equal bow height error to generate a set of discrete pose points with timestamps, superimpose the AGV dynamics constraint to generate a speed look-ahead control instruction set, split the instruction set into independent control units according to the process section, encapsulate it into time-triggered messages through the industrial bus protocol, and inject it into the physical execution unit based on the simulation clock synchronization signal.

2. The aluminum die plate production line discrete event simulation intelligent control method according to claim 1, characterized in that, Collect welding current pulse waveform, infrared temperature field distribution and vibration spectrum data; According to the pulse-temperature-vibration data stream, generate a dynamic event package with a process label; Inject the dynamic event package into the priority queue of the discrete event simulation engine, including: Step 11, by synchronously collecting welding current pulse waveform, infrared temperature field distribution and vibration spectrum data, align the timestamps of the three data sets according to a unified clock source to generate a three-dimensional sensing signal set; Step 12, based on the three-dimensional sensing signal set, extract current waveform variation features, temperature thermal field gradient features and vibration energy accumulation features, match the process type identifier according to the feature combination, and generate a dynamic event entity; Step 13, based on the dynamic event entity, analyze the process type identifier to determine the basic priority level, superimpose the feature intensity parameters to calculate the final priority, load the event entity into the priority queue of the discrete event simulation engine in descending order of priority, and clear the historical event entity with the lowest priority when the queue is full.

3. The aluminum die plate production line discrete event simulation intelligent control method according to claim 2, characterized in that, The step 2 comprises: Step 21, extract the dynamic event package from the priority queue to generate a feature analysis result containing the process label and temperature field distribution data; based on the temperature field distribution data in the feature analysis result, calculate the thermal accumulation gradient change rate between adjacent thermal analysis nodes in the welding area to obtain a gradient change rate matrix; Step 22, according to the gradient change rate matrix and the process label, match the pose compensation rules in the process knowledge base, and output a compensation rule instruction set; based on the compensation rule instruction set, generate a six-degree-of-freedom pose compensation vector of the welding robot twin. Step 23, decompose the pose compensation vector to get decoupled translation and rotation components, perform position compensation based on the translation component to generate compensated position coordinates, and perform pose fusion based on the rotation component to generate compensated pose data; based on the compensated pose data, check joint limits and generate an out-of-limit state flag through inverse kinematics; Step 24, based on the state flag and the pose compensation vector, scale the compensation to generate the final pose parameters, and write the final pose parameters into the kinematics model database to generate updated kinematics parameters.

4. The aluminum die plate production line discrete event simulation intelligent control method according to claim 3, characterized in that, Based on the updated kinematics model parameters, advance the discrete event simulation clock; if the simulation event triggers the logistics decision node, suspend the event scheduler and intercept the thermal constraint state matrix; Use the DQN algorithm to generate a dynamic disturbance-resistant scheduling strategy, including: Step 31, based on the updated kinematics model parameters, advance the discrete event simulation clock to generate the current simulation event set; Step 32, receive the simulation event set and parse the event type identifier and device state parameters, generate a decision trigger flag based on the event type identifier matching the logistics decision node trigger condition, and respond to the flag to suspend the event scheduler and lock the simulation clock timestamp; Step 33, based on the locked timestamp, intercept the temperature field distribution matrix, thermal accumulation gradient vector, and device cooling state flag from the digital twin database, and package the thermal constraint state data into a structured thermal constraint state matrix; Step 34, receive the thermal constraint state matrix, perform standardization preprocessing to generate an input state vector; input the input state vector into the policy network, calculate the action value evaluation matrix through multiple layers of neurons, and perform value-guided decision-making based on the action value evaluation matrix to generate a dynamic disturbance-resistant scheduling strategy.

5. The aluminum die plate production line discrete event simulation intelligent control method according to claim 4, characterized in that, Parse the material transfer path instructions and cutting process key point coordinate sequence from the dynamic disturbance-resistant scheduling strategy, and input the key point coordinate sequence into the curve driving algorithm to generate a parameterized continuous cutting path curve based on the aluminum template contour accuracy requirements and cutting head dynamics constraints, including: Step 41, parse the dynamic disturbance-resistant scheduling strategy to get the material transfer path topology sequence, and separate the cutting process key point coordinate data set; Step 42, based on the key point coordinate data set, construct the path control structure in the order of the process; in the path control structure, connect the key points using the parametric spline algorithm to get the initial path curve; scan and extract the contour feature points along the initial path curve, identify the curvature mutation area; according to the aluminum template contour accuracy constraint, apply smoothing processing in the curvature mutation area to generate an optimized path curve; insert a curvature continuous transition section at the path curve junction to get an industrial-grade smooth path curve; Step 43, receive the industrial-grade smooth path curve, parse the cutting head dynamics parameter constraints, calculate the acceleration sensitive area along the path curve, and insert a velocity smoothing section according to the dynamics constraints to generate a position-velocity-acceleration three-order coupled parameterized continuous cutting path curve.

6. The aluminum die plate production line discrete event simulation intelligent control method according to claim 5, characterized in that, The continuous cutting path curve is discretized with equal error to generate a discrete pose point set with time stamp, the AGV dynamics constraint is superimposed to generate a speed look-ahead control instruction set, the instruction set is split into independent control units according to the process section, the control units are packaged into time-triggered messages through the industrial bus protocol, and the physical execution unit is injected based on the simulation clock synchronization signal, including: Step 51, receiving a parameterized continuous cutting path curve, discretizing sampling based on equal error constraint, and generating a discrete pose point set; Step 52, receiving the discrete pose point set, analyzing the space coordinates and timestamp information in the pose point, calculating the path curvature distribution based on the space coordinates, and identifying the high dynamic constraint section; the speed profile is generated by applying AGV dynamics constraint in the high dynamic constraint section through the speed look-ahead algorithm; the speed profile and the pose point set are fused to generate a motion control instruction set; Step 53, the motion control instruction set is split into independent control units according to the process characteristics; the control units are packaged into time-triggered messages through the industrial bus protocol; the time-triggered messages are injected into the physical execution unit based on the discrete event simulation clock synchronization signal.

7. The aluminum die plate production line discrete event simulation intelligent control method according to claim 6, characterized in that, Receiving the discrete pose point set, analyzing the space coordinates and timestamp information in the pose point, calculating the path curvature distribution based on the space coordinates, and identifying the high dynamic constraint section; the speed profile is generated by applying AGV dynamics constraint in the high dynamic constraint section through the speed look-ahead algorithm; Fusing the speed profile and the pose point set to generate a motion control instruction set, including: Step 61, based on the space coordinate information of the discrete pose point set, the instantaneous curvature value of the path segment is calculated, the curvature distribution sequence is generated, the high dynamic feature points are identified by comparing the preset curvature threshold, and the continuous high dynamic feature points are combined to generate a high dynamic constraint section; Step 62, based on the high dynamic constraint section, the maximum centripetal acceleration parameter of the AGV is combined to calculate the upper limit of the safe speed of each section, the allowed speed value between adjacent pose points is iteratively calculated from the end point to the starting point of the section, and the speed decreasing sequence is generated by applying the maximum acceleration and deceleration constraint of the AGV, and the uniform speed profile is fused to generate the speed profile; Step 63, the speed value of the speed profile is bound to the pose point, the space coordinates, attitude angle, speed value and timestamp are combined to generate a structured instruction unit, and the motion control instruction set is generated after being arranged in ascending order of timestamp.

8. An intelligent control system for discrete event simulation of an aluminum die plate production line, the system implementing the method of any one of claims 1 to 7, characterized in that, Including: The acquisition module is used for collecting welding current pulse waveform, infrared temperature field distribution and vibration spectrum data; According to the pulse-temperature-vibration data stream, a dynamic event package with a process label is generated; The dynamic event package is injected into the priority queue of the discrete event simulation engine; The calculation module is used for analyzing the thermal accumulation gradient parameters in the feature vector based on the dynamic event package in the priority queue; Based on the thermal accumulation gradient parameters, the pose compensation amount of the welding robot twin is calculated, and the kinematic model parameters of the twin are updated; The execution module is used for advancing the discrete event simulation clock based on the updated kinematic model parameters; if the simulation event triggers the logistics decision node, the event scheduler is suspended and the thermal constraint state matrix is intercepted; The thermal constraint state matrix is used to generate a dynamic anti-disturbance scheduling strategy by using the DQN algorithm; A constraint module is configured to parse material transfer path instructions and cutting process key point coordinate sequences from a dynamic anti-disturbance scheduling strategy, input the key point coordinate sequences into a curve driving algorithm, and generate a parameterized continuous cutting path curve according to aluminum template contour accuracy requirements and cutting head dynamics constraints; A processing module is configured to perform equal-bow-height-error discretization processing on the continuous cutting path curve to generate a discrete pose point set with timestamps, superimpose AGV dynamics constraints to generate a speed look-ahead control instruction set, split the instruction set into independent control units according to process sections, encapsulate the instruction set into time-triggered messages through an industrial bus protocol, and inject a physical execution unit based on a simulation clock synchronization signal.

9. A computing device, comprising: comprise: one or more processors; a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a program which, when executed by a processor, implements the method of any one of claims 1 to 7.

Images